You don't have to keep up

I needed a series of diagrams for a new site build I’ve recently been working on. My first thought was to go find a tool that makes infographics. How about, NotebookLM?

That reflex is, in many ways, the whole problem. New output, so a new tool. It’s the same move that has many of us feeling overwhelmed.

What changed my mind was a small shift in the question. Instead of asking what makes infographics, I asked what do I actually want as the outcome.

The tool I already had drew them

What I wanted was 6 visuals that felt like one set and looked like they belonged on that site. Same fonts. Same colours. The same slightly hand-drawn elements that show up everywhere else on the pages.

NotebookLM can make a fine infographic, and to be fair, it’s a good tool. What it can’t do is make one in my typefaces, on my colour grounds, with my style of hand drawn elements. And, it can’t be edited with a sentence a month from now.

So the diagrams got built where I was already working. Plain SVG, hand-built by Claude Code in the site’s own design system, all of it living in a single component called ProjectDiagram.astro. No diagramming tool involved.

The interesting part, the result was better than the specialist tool would have given me. Not adequate, better.

The tool I already had, drew them. That’s the point.

The race is rigged

There’s something new in AI every day (let’s be realistic, multiple times a day!), and each launch arrives framed as the one you can’t afford to miss. Miss enough of them and you fear falling behind.

I’d like to say something plainly. It’s a feature of the launch cycle, not a measurement of you. The cycle is built to produce it. Treating it as information about your own progress is a mistake, and an expensive one, because the response it pushes you toward, sampling a bit of everything, is the response least likely to make you better at any of it.

Here’s what made me stop chasing.

The way you work with a system compounds. How you frame a problem. What context you hand over before you ask for anything. When you step in and when you let it run. What you keep from the last time. None of that belongs to a product, and all of it gets better with use.

That’s why going deep on one is faster than sampling 12. You’re not learning the tool. You’re learning to think alongside it, and that skill walks with you to whatever comes next, because the next tool is a different door into a skill you already have.

What actually travels

The clearest evidence I have for that is boring on purpose.

I keep 3 foundation files. One holds my voice, one holds who I am and what I care about, and one holds the business. They’re plain Markdown. I wrote them once, and only update when something material changes (read these next 3 words carefully), in my world.

They work in whatever I open next. Nothing in them belongs to a product. When something new shows up and I want to try it seriously, I don’t start from zero, I hand it those files and it knows who it’s working with.

So the door line isn’t a metaphor. It’s how those files literally function. The part that took real effort, the context and the judgment, is portable by construction. The tools are interchangeable in a way the context never is.

Don’t chase, go deep.

Where I’d start

If any of this sounds like your current relationship with AI, this is what I’d actually do, in order.

Pick one foundation model and stay there a while. ChatGPT, Claude, Gemini, the choice matters less than staying put. Learning to communicate with one of them is the gateway skill. Prompting isn’t a trick you learn per product, it’s the practice of being specific about what you want and what good looks like, and being specific travels everywhere.

Find one real pain point and augment it. Automate is the word that usually shows up here, and augment is the more honest one, since the goal was never to remove yourself from the work. So… not your whole job, one thing. Something you do often, or don’t enjoy. Drafting the same kind of email, getting a rough outline started, making sense of your own notes. Solving one gives you something in addition to the time back. It gives you confidence, and confidence is what actually ends the overwhelm. Anxiety doesn’t respond to reading about AI. It responds to having done something with it.

Build systems, not a collection of apps. You don’t need to become a programmer, and you don’t need to learn every app that launches. You need reusable things. Custom instructions. A few templates. Targeted agents, and targeted is the key word there, meaning it does one specific job every time rather than being a general assistant you have to re-explain yourself to. Something that keeps working after you close the tab.

The first 2 make you capable. The third is what makes the tools disposable. An app you learned is something you’ll have to learn again. A system you built is something you keep.

You’re not behind

If you’re feeling behind, I don’t think you are. I think you’re early in a skill that takes longer to build than a launch cycle takes to run, and those two things get confused constantly.

The shiny new object will keep arriving on schedule. It’s fine to let most of it go by.

What have you stopped chasing, and did it cost you anything?